Fig. 1

Performance curves for DTU analysis based on two simulation studies. The false discovery proportion (FDP, x-axis) is the fraction of false positive hypotheses over all rejected hypotheses. The true positive rate (TPR, y-axis) represents the fraction of false null hypotheses that have indeed been rejected. The three points on each curve represent working points on a nominal 1%, 5% and 10% FDR. The left panel a shows the results from a simulation performed in Soneson et al. (2016) [24] based on the Drosophila melanogaster transcriptome and clearly shows the increased sensitivity for tests that aggregate all transcript hypotheses on a gene level (green curve) in comparison to transcript-level tests (blue curve). The right panel b shows the results from a simulation based on the human transcriptome used in Soneson et al. (2016) [10]. Here, aggregated hypothesis tests show an even larger increase in sensitivity, possibly due to the higher complexity of the human transcriptome and thus a higher expected number of transcripts per gene for human